diff --git a/src/cpu/o3/commit.hh b/src/cpu/o3/commit.hh
index bf0b07ca29..4c9a7b5a2d 100644
--- a/src/cpu/o3/commit.hh
+++ b/src/cpu/o3/commit.hh
@@ -87,7 +87,6 @@ class DefaultCommit
 {
   public:
     // Typedefs from the Impl.
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::TimeStruct TimeStruct;
     typedef typename Impl::FetchStruct FetchStruct;
     typedef typename Impl::IEWStruct IEWStruct;
@@ -136,7 +135,7 @@ class DefaultCommit
 
   public:
     /** Construct a DefaultCommit with the given parameters. */
-    DefaultCommit(O3CPU *_cpu, const DerivO3CPUParams &params);
+    DefaultCommit(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     /** Returns the name of the DefaultCommit. */
     std::string name() const;
@@ -357,7 +356,7 @@ class DefaultCommit
 
   private:
     /** Pointer to O3CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Vector of all of the threads. */
     std::vector<Thread *> thread;
@@ -480,7 +479,7 @@ class DefaultCommit
 
     struct CommitStats : public Stats::Group
     {
-        CommitStats(O3CPU *cpu, DefaultCommit *commit);
+        CommitStats(FullO3CPU<Impl> *cpu, DefaultCommit *commit);
         /** Stat for the total number of squashed instructions discarded by
          * commit.
          */
diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh
index 06694bf257..4442eb9904 100644
--- a/src/cpu/o3/commit_impl.hh
+++ b/src/cpu/o3/commit_impl.hh
@@ -79,7 +79,8 @@ DefaultCommit<Impl>::processTrapEvent(ThreadID tid)
 }
 
 template <class Impl>
-DefaultCommit<Impl>::DefaultCommit(O3CPU *_cpu, const DerivO3CPUParams &params)
+DefaultCommit<Impl>::DefaultCommit(FullO3CPU<Impl> *_cpu,
+        const DerivO3CPUParams &params)
     : commitPolicy(params.smtCommitPolicy),
       cpu(_cpu),
       iewToCommitDelay(params.iewToCommitDelay),
@@ -150,7 +151,7 @@ DefaultCommit<Impl>::regProbePoints()
 }
 
 template <class Impl>
-DefaultCommit<Impl>::CommitStats::CommitStats(O3CPU *cpu,
+DefaultCommit<Impl>::CommitStats::CommitStats(FullO3CPU<Impl> *cpu,
                                               DefaultCommit *commit)
     : Stats::Group(cpu, "commit"),
       ADD_STAT(commitSquashedInsts, Stats::Units::Count::get(),
@@ -344,7 +345,7 @@ DefaultCommit<Impl>::startupStage()
 
     // Commit must broadcast the number of free entries it has at the
     // start of the simulation, so it starts as active.
-    cpu->activateStage(O3CPU::CommitIdx);
+    cpu->activateStage(FullO3CPU<Impl>::CommitIdx);
 
     cpu->activityThisCycle();
 }
@@ -496,10 +497,10 @@ DefaultCommit<Impl>::updateStatus()
 
     if (_nextStatus == Inactive && _status == Active) {
         DPRINTF(Activity, "Deactivating stage.\n");
-        cpu->deactivateStage(O3CPU::CommitIdx);
+        cpu->deactivateStage(FullO3CPU<Impl>::CommitIdx);
     } else if (_nextStatus == Active && _status == Inactive) {
         DPRINTF(Activity, "Activating stage.\n");
-        cpu->activateStage(O3CPU::CommitIdx);
+        cpu->activateStage(FullO3CPU<Impl>::CommitIdx);
     }
 
     _status = _nextStatus;
diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc
index 1ed725b8ee..07e0893709 100644
--- a/src/cpu/o3/cpu.cc
+++ b/src/cpu/o3/cpu.cc
@@ -315,7 +315,7 @@ FullO3CPU<Impl>::FullO3CPU(const DerivO3CPUParams &params)
                 DPRINTF(O3CPU, "Workload[%i] process is %#x",
                         tid, this->thread[tid]);
                 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
-                        (typename Impl::O3CPU *)(this),
+                        (FullO3CPU *)(this),
                         tid, params.workload[tid]);
 
                 //usedTids[tid] = true;
@@ -326,8 +326,7 @@ FullO3CPU<Impl>::FullO3CPU(const DerivO3CPUParams &params)
                 Process* dummy_proc = NULL;
 
                 this->thread[tid] = new typename FullO3CPU<Impl>::Thread(
-                        (typename Impl::O3CPU *)(this),
-                        tid, dummy_proc);
+                        this, tid, dummy_proc);
                 //usedTids[tid] = false;
             }
         }
@@ -346,8 +345,7 @@ FullO3CPU<Impl>::FullO3CPU(const DerivO3CPUParams &params)
                 o3_tc, this->checker);
         }
 
-        o3_tc->cpu = (typename Impl::O3CPU *)(this);
-        assert(o3_tc->cpu);
+        o3_tc->cpu = this;
         o3_tc->thread = this->thread[tid];
 
         // Give the thread the TC.
@@ -391,8 +389,7 @@ FullO3CPU<Impl>::regProbePoints()
 }
 
 template <class Impl>
-FullO3CPU<Impl>::
-FullO3CPUStats::FullO3CPUStats(FullO3CPU *cpu)
+FullO3CPU<Impl>::FullO3CPUStats::FullO3CPUStats(FullO3CPU *cpu)
     : Stats::Group(cpu),
       ADD_STAT(timesIdled, Stats::Units::Count::get(),
                "Number of times that the entire CPU went into an idle state "
diff --git a/src/cpu/o3/cpu.hh b/src/cpu/o3/cpu.hh
index fbf402e9d9..2eeedea004 100644
--- a/src/cpu/o3/cpu.hh
+++ b/src/cpu/o3/cpu.hh
@@ -101,8 +101,6 @@ class FullO3CPU : public BaseO3CPU
 {
   public:
     // Typedefs from the Impl here.
-    typedef typename Impl::O3CPU O3CPU;
-
     typedef O3ThreadState<Impl> ImplState;
     typedef O3ThreadState<Impl> Thread;
 
diff --git a/src/cpu/o3/decode.hh b/src/cpu/o3/decode.hh
index 38ba0a6cd3..1d9e04e1fd 100644
--- a/src/cpu/o3/decode.hh
+++ b/src/cpu/o3/decode.hh
@@ -50,6 +50,9 @@
 
 struct DerivO3CPUParams;
 
+template <class Impl>
+class FullO3CPU;
+
 /**
  * DefaultDecode class handles both single threaded and SMT
  * decode. Its width is specified by the parameters; each cycles it
@@ -62,7 +65,6 @@ class DefaultDecode
 {
   private:
     // Typedefs from the Impl.
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::FetchStruct FetchStruct;
     typedef typename Impl::DecodeStruct DecodeStruct;
     typedef typename Impl::TimeStruct TimeStruct;
@@ -97,7 +99,7 @@ class DefaultDecode
 
   public:
     /** DefaultDecode constructor. */
-    DefaultDecode(O3CPU *_cpu, const DerivO3CPUParams &params);
+    DefaultDecode(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     void startupStage();
 
@@ -204,7 +206,7 @@ class DefaultDecode
   private:
     // Interfaces to objects outside of decode.
     /** CPU interface. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Time buffer interface. */
     TimeBuffer<TimeStruct> *timeBuffer;
@@ -295,7 +297,7 @@ class DefaultDecode
 
     struct DecodeStats : public Stats::Group
     {
-        DecodeStats(O3CPU *cpu);
+        DecodeStats(FullO3CPU<Impl> *cpu);
 
         /** Stat for total number of idle cycles. */
         Stats::Scalar idleCycles;
diff --git a/src/cpu/o3/decode_impl.hh b/src/cpu/o3/decode_impl.hh
index bfbf98f67b..fc03145bb9 100644
--- a/src/cpu/o3/decode_impl.hh
+++ b/src/cpu/o3/decode_impl.hh
@@ -59,7 +59,8 @@
 using std::list;
 
 template<class Impl>
-DefaultDecode<Impl>::DefaultDecode(O3CPU *_cpu, const DerivO3CPUParams &params)
+DefaultDecode<Impl>::DefaultDecode(FullO3CPU<Impl> *_cpu,
+        const DerivO3CPUParams &params)
     : cpu(_cpu),
       renameToDecodeDelay(params.renameToDecodeDelay),
       iewToDecodeDelay(params.iewToDecodeDelay),
@@ -122,7 +123,7 @@ DefaultDecode<Impl>::name() const
 }
 
 template <class Impl>
-DefaultDecode<Impl>::DecodeStats::DecodeStats(O3CPU *cpu)
+DefaultDecode<Impl>::DecodeStats::DecodeStats(FullO3CPU<Impl> *cpu)
     : Stats::Group(cpu, "decode"),
       ADD_STAT(idleCycles, Stats::Units::Cycle::get(),
                "Number of cycles decode is idle"),
@@ -457,7 +458,7 @@ DefaultDecode<Impl>::updateStatus()
 
             DPRINTF(Activity, "Activating stage.\n");
 
-            cpu->activateStage(O3CPU::DecodeIdx);
+            cpu->activateStage(FullO3CPU<Impl>::DecodeIdx);
         }
     } else {
         // If it's not unblocking, then decode will not have any internal
@@ -466,7 +467,7 @@ DefaultDecode<Impl>::updateStatus()
             _status = Inactive;
             DPRINTF(Activity, "Deactivating stage.\n");
 
-            cpu->deactivateStage(O3CPU::DecodeIdx);
+            cpu->deactivateStage(FullO3CPU<Impl>::DecodeIdx);
         }
     }
 }
diff --git a/src/cpu/o3/fetch.hh b/src/cpu/o3/fetch.hh
index dee344b414..2b9e2de84a 100644
--- a/src/cpu/o3/fetch.hh
+++ b/src/cpu/o3/fetch.hh
@@ -73,7 +73,6 @@ class DefaultFetch
 {
   public:
     /** Typedefs from Impl. */
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::FetchStruct FetchStruct;
     typedef typename Impl::TimeStruct TimeStruct;
 
@@ -212,7 +211,7 @@ class DefaultFetch
 
   public:
     /** DefaultFetch constructor. */
-    DefaultFetch(O3CPU *_cpu, const DerivO3CPUParams &params);
+    DefaultFetch(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     /** Returns the name of fetch. */
     std::string name() const;
@@ -402,7 +401,7 @@ class DefaultFetch
 
   private:
     /** Pointer to the O3CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Time buffer interface. */
     TimeBuffer<TimeStruct> *timeBuffer;
@@ -544,7 +543,7 @@ class DefaultFetch
   protected:
     struct FetchStatGroup : public Stats::Group
     {
-        FetchStatGroup(O3CPU *cpu, DefaultFetch *fetch);
+        FetchStatGroup(FullO3CPU<Impl> *cpu, DefaultFetch *fetch);
         // @todo: Consider making these
         // vectors and tracking on a per thread basis.
         /** Stat for total number of cycles stalled due to an icache miss. */
diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh
index 587ae1ae18..a5627e1644 100644
--- a/src/cpu/o3/fetch_impl.hh
+++ b/src/cpu/o3/fetch_impl.hh
@@ -73,7 +73,8 @@
 #include "sim/system.hh"
 
 template<class Impl>
-DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, const DerivO3CPUParams &params)
+DefaultFetch<Impl>::DefaultFetch(FullO3CPU<Impl> *_cpu,
+        const DerivO3CPUParams &params)
     : fetchPolicy(params.smtFetchPolicy),
       cpu(_cpu),
       branchPred(nullptr),
@@ -158,7 +159,7 @@ DefaultFetch<Impl>::regProbePoints()
 
 template <class Impl>
 DefaultFetch<Impl>::
-FetchStatGroup::FetchStatGroup(O3CPU *cpu, DefaultFetch *fetch)
+FetchStatGroup::FetchStatGroup(FullO3CPU<Impl> *cpu, DefaultFetch *fetch)
     : Stats::Group(cpu, "fetch"),
     ADD_STAT(icacheStallCycles, Stats::Units::Cycle::get(),
              "Number of cycles fetch is stalled on an Icache miss"),
@@ -493,7 +494,7 @@ DefaultFetch<Impl>::switchToActive()
     if (_status == Inactive) {
         DPRINTF(Activity, "Activating stage.\n");
 
-        cpu->activateStage(O3CPU::FetchIdx);
+        cpu->activateStage(FullO3CPU<Impl>::FetchIdx);
 
         _status = Active;
     }
@@ -506,7 +507,7 @@ DefaultFetch<Impl>::switchToInactive()
     if (_status == Active) {
         DPRINTF(Activity, "Deactivating stage.\n");
 
-        cpu->deactivateStage(O3CPU::FetchIdx);
+        cpu->deactivateStage(FullO3CPU<Impl>::FetchIdx);
 
         _status = Inactive;
     }
@@ -831,7 +832,7 @@ DefaultFetch<Impl>::updateFetchStatus()
                             "completion\n",tid);
                 }
 
-                cpu->activateStage(O3CPU::FetchIdx);
+                cpu->activateStage(FullO3CPU<Impl>::FetchIdx);
             }
 
             return Active;
@@ -842,7 +843,7 @@ DefaultFetch<Impl>::updateFetchStatus()
     if (_status == Active) {
         DPRINTF(Activity, "Deactivating stage.\n");
 
-        cpu->deactivateStage(O3CPU::FetchIdx);
+        cpu->deactivateStage(FullO3CPU<Impl>::FetchIdx);
     }
 
     return Inactive;
diff --git a/src/cpu/o3/iew.hh b/src/cpu/o3/iew.hh
index 4afee5bf76..a0f48697f0 100644
--- a/src/cpu/o3/iew.hh
+++ b/src/cpu/o3/iew.hh
@@ -82,7 +82,6 @@ class DefaultIEW
 {
   private:
     //Typedefs from Impl
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::TimeStruct TimeStruct;
     typedef typename Impl::IEWStruct IEWStruct;
     typedef typename Impl::RenameStruct RenameStruct;
@@ -129,7 +128,7 @@ class DefaultIEW
 
   public:
     /** Constructs a DefaultIEW with the given parameters. */
-    DefaultIEW(O3CPU *_cpu, const DerivO3CPUParams &params);
+    DefaultIEW(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     /** Returns the name of the DefaultIEW stage. */
     std::string name() const;
@@ -347,7 +346,7 @@ class DefaultIEW
 
   private:
     /** CPU pointer. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Records if IEW has written to the time buffer this cycle, so that the
      * CPU can deschedule itself if there is no activity.
@@ -424,7 +423,7 @@ class DefaultIEW
 
     struct IEWStats : public Stats::Group
     {
-        IEWStats(O3CPU *cpu);
+        IEWStats(FullO3CPU<Impl> *cpu);
 
         /** Stat for total number of idle cycles. */
         Stats::Scalar idleCycles;
@@ -460,7 +459,7 @@ class DefaultIEW
 
         struct ExecutedInstStats : public Stats::Group
         {
-            ExecutedInstStats(O3CPU* cpu);
+            ExecutedInstStats(FullO3CPU<Impl> *cpu);
 
             /** Stat for total number of executed instructions. */
             Stats::Scalar numInsts;
diff --git a/src/cpu/o3/iew_impl.hh b/src/cpu/o3/iew_impl.hh
index 7c6fe5a6a9..54327fa5a4 100644
--- a/src/cpu/o3/iew_impl.hh
+++ b/src/cpu/o3/iew_impl.hh
@@ -62,7 +62,8 @@
 #include "params/DerivO3CPU.hh"
 
 template<class Impl>
-DefaultIEW<Impl>::DefaultIEW(O3CPU *_cpu, const DerivO3CPUParams &params)
+DefaultIEW<Impl>::DefaultIEW(FullO3CPU<Impl> *_cpu,
+        const DerivO3CPUParams &params)
     : issueToExecQueue(params.backComSize, params.forwardComSize),
       cpu(_cpu),
       instQueue(_cpu, this, params),
@@ -142,8 +143,7 @@ DefaultIEW<Impl>::regProbePoints()
 }
 
 template <class Impl>
-DefaultIEW<Impl>::
-IEWStats::IEWStats(O3CPU *cpu)
+DefaultIEW<Impl>::IEWStats::IEWStats(FullO3CPU<Impl> *cpu)
     : Stats::Group(cpu),
     ADD_STAT(idleCycles, Stats::Units::Cycle::get(),
              "Number of cycles IEW is idle"),
@@ -218,8 +218,8 @@ IEWStats::IEWStats(O3CPU *cpu)
 }
 
 template <class Impl>
-DefaultIEW<Impl>::IEWStats::
-ExecutedInstStats::ExecutedInstStats(O3CPU *cpu)
+DefaultIEW<Impl>::IEWStats::ExecutedInstStats::ExecutedInstStats(
+        FullO3CPU<Impl> *cpu)
     : Stats::Group(cpu),
     ADD_STAT(numInsts, Stats::Units::Count::get(),
              "Number of executed instructions"),
@@ -288,7 +288,7 @@ DefaultIEW<Impl>::startupStage()
         cpu->checker->setDcachePort(&ldstQueue.getDataPort());
     }
 
-    cpu->activateStage(O3CPU::IEWIdx);
+    cpu->activateStage(FullO3CPU<Impl>::IEWIdx);
 }
 
 template<class Impl>
@@ -865,7 +865,7 @@ inline void
 DefaultIEW<Impl>::activateStage()
 {
     DPRINTF(Activity, "Activating stage.\n");
-    cpu->activateStage(O3CPU::IEWIdx);
+    cpu->activateStage(FullO3CPU<Impl>::IEWIdx);
 }
 
 template <class Impl>
@@ -873,7 +873,7 @@ inline void
 DefaultIEW<Impl>::deactivateStage()
 {
     DPRINTF(Activity, "Deactivating stage.\n");
-    cpu->deactivateStage(O3CPU::IEWIdx);
+    cpu->deactivateStage(FullO3CPU<Impl>::IEWIdx);
 }
 
 template<class Impl>
diff --git a/src/cpu/o3/impl.hh b/src/cpu/o3/impl.hh
index 2c7242ea60..d9d3e1fc82 100644
--- a/src/cpu/o3/impl.hh
+++ b/src/cpu/o3/impl.hh
@@ -31,10 +31,6 @@
 
 #include "cpu/o3/comm.hh"
 
-// Forward declarations.
-template <class Impl>
-class FullO3CPU;
-
 /** Implementation specific struct that defines several key types to the
  *  CPU, the stages within the CPU, the time buffers, and the DynInst.
  *  The struct defines the ISA, the CPU policy, the specific DynInst, the
@@ -62,16 +58,6 @@ struct O3CPUImpl
 
     /** The struct for all backwards communication. */
     typedef TimeBufStruct<O3CPUImpl> TimeStruct;
-
-
-    /** The O3CPU type to be used. */
-    typedef FullO3CPU<O3CPUImpl> O3CPU;
-
-    /** Same typedef, but for CPUType.  BaseDynInst may not always use
-     * an O3 CPU, so it's clearer to call it CPUType instead in that
-     * case.
-     */
-    typedef O3CPU CPUType;
 };
 
 #endif // __CPU_O3_SPARC_IMPL_HH__
diff --git a/src/cpu/o3/inst_queue.hh b/src/cpu/o3/inst_queue.hh
index 6c85ffd87e..879655d6be 100644
--- a/src/cpu/o3/inst_queue.hh
+++ b/src/cpu/o3/inst_queue.hh
@@ -67,6 +67,9 @@ class MemInterface;
 template <class Impl>
 class DefaultIEW;
 
+template <class Impl>
+class FullO3CPU;
+
 /**
  * A standard instruction queue class.  It holds ready instructions, in
  * order, in seperate priority queues to facilitate the scheduling of
@@ -89,7 +92,6 @@ class InstructionQueue
 {
   public:
     //Typedefs from the Impl.
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::IssueStruct IssueStruct;
     typedef typename Impl::TimeStruct TimeStruct;
 
@@ -125,7 +127,7 @@ class InstructionQueue
     };
 
     /** Constructs an IQ. */
-    InstructionQueue(O3CPU *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
+    InstructionQueue(FullO3CPU<Impl> *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
                      const DerivO3CPUParams &params);
 
     /** Destructs the IQ. */
@@ -282,7 +284,7 @@ class InstructionQueue
     /////////////////////////
 
     /** Pointer to the CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Cache interface. */
     MemInterface *dcacheInterface;
@@ -479,7 +481,7 @@ class InstructionQueue
 
     struct IQStats : public Stats::Group
     {
-        IQStats(O3CPU *cpu, const unsigned &total_width);
+        IQStats(FullO3CPU<Impl> *cpu, const unsigned &total_width);
         /** Stat for number of instructions added. */
         Stats::Scalar instsAdded;
         /** Stat for number of non-speculative instructions added. */
diff --git a/src/cpu/o3/inst_queue_impl.hh b/src/cpu/o3/inst_queue_impl.hh
index 9373f6b879..315974247e 100644
--- a/src/cpu/o3/inst_queue_impl.hh
+++ b/src/cpu/o3/inst_queue_impl.hh
@@ -84,7 +84,7 @@ InstructionQueue<Impl>::FUCompletion::description() const
 }
 
 template <class Impl>
-InstructionQueue<Impl>::InstructionQueue(O3CPU *cpu_ptr,
+InstructionQueue<Impl>::InstructionQueue(FullO3CPU<Impl> *cpu_ptr,
         DefaultIEW<Impl> *iew_ptr, const DerivO3CPUParams &params)
     : cpu(cpu_ptr),
       iewStage(iew_ptr),
@@ -177,7 +177,7 @@ InstructionQueue<Impl>::name() const
 
 template <class Impl>
 InstructionQueue<Impl>::
-IQStats::IQStats(O3CPU *cpu, const unsigned &total_width)
+IQStats::IQStats(FullO3CPU<Impl> *cpu, const unsigned &total_width)
     : Stats::Group(cpu),
     ADD_STAT(instsAdded, Stats::Units::Count::get(),
              "Number of instructions added to the IQ (excludes non-spec)"),
diff --git a/src/cpu/o3/lsq.hh b/src/cpu/o3/lsq.hh
index eb76e655a3..021bab5380 100644
--- a/src/cpu/o3/lsq.hh
+++ b/src/cpu/o3/lsq.hh
@@ -75,8 +75,6 @@ template <class Impl>
 class LSQ
 {
   public:
-    typedef typename Impl::O3CPU O3CPU;
-
     class LSQRequest;
     /** Derived class to hold any sender state the LSQ needs. */
     class LSQSenderState : public Packet::SenderState
@@ -794,7 +792,7 @@ class LSQ
     };
 
     /** Constructs an LSQ with the given parameters. */
-    LSQ(O3CPU *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
+    LSQ(FullO3CPU<Impl> *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
             const DerivO3CPUParams &params);
     ~LSQ() { }
 
@@ -1051,7 +1049,7 @@ class LSQ
                       const std::vector<bool>& byte_enable);
 
     /** The CPU pointer. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** The IEW stage pointer. */
     DefaultIEW<Impl> *iewStage;
diff --git a/src/cpu/o3/lsq_impl.hh b/src/cpu/o3/lsq_impl.hh
index 5cde78e0f8..bb53bb62fc 100644
--- a/src/cpu/o3/lsq_impl.hh
+++ b/src/cpu/o3/lsq_impl.hh
@@ -68,7 +68,7 @@ LSQ<Impl>::LSQSenderState::contextId()
 }
 
 template <class Impl>
-LSQ<Impl>::LSQ(O3CPU *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
+LSQ<Impl>::LSQ(FullO3CPU<Impl> *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
         const DerivO3CPUParams &params)
     : cpu(cpu_ptr), iewStage(iew_ptr),
       _cacheBlocked(false),
diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh
index 4d3c41958e..14768f8d20 100644
--- a/src/cpu/o3/lsq_unit.hh
+++ b/src/cpu/o3/lsq_unit.hh
@@ -85,7 +85,6 @@ class LSQUnit
   public:
     static constexpr auto MaxDataBytes = MaxVecRegLenInBytes;
 
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::IssueStruct IssueStruct;
 
     using LSQSenderState = typename LSQ<Impl>::LSQSenderState;
@@ -225,7 +224,7 @@ class LSQUnit
     }
 
     /** Initializes the LSQ unit with the specified number of entries. */
-    void init(O3CPU *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
+    void init(FullO3CPU<Impl> *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
             const DerivO3CPUParams &params, LSQ<Impl> *lsq_ptr, unsigned id);
 
     /** Returns the name of the LSQ unit. */
@@ -398,7 +397,7 @@ class LSQUnit
 
   private:
     /** Pointer to the CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Pointer to the IEW stage. */
     DefaultIEW<Impl> *iewStage;
diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh
index 174916df99..df789ce6b3 100644
--- a/src/cpu/o3/lsq_unit_impl.hh
+++ b/src/cpu/o3/lsq_unit_impl.hh
@@ -215,7 +215,7 @@ LSQUnit<Impl>::LSQUnit(uint32_t lqEntries, uint32_t sqEntries)
 
 template<class Impl>
 void
-LSQUnit<Impl>::init(O3CPU *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
+LSQUnit<Impl>::init(FullO3CPU<Impl> *cpu_ptr, DefaultIEW<Impl> *iew_ptr,
         const DerivO3CPUParams &params, LSQ<Impl> *lsq_ptr, unsigned id)
 {
     lsqID = id;
diff --git a/src/cpu/o3/mem_dep_unit.hh b/src/cpu/o3/mem_dep_unit.hh
index b2cf9bf8d2..3874e5fa32 100644
--- a/src/cpu/o3/mem_dep_unit.hh
+++ b/src/cpu/o3/mem_dep_unit.hh
@@ -68,6 +68,9 @@ struct DerivO3CPUParams;
 template <class Impl>
 class InstructionQueue;
 
+template <class Impl>
+class FullO3CPU;
+
 /**
  * Memory dependency unit class.  This holds the memory dependence predictor.
  * As memory operations are issued to the IQ, they are also issued to this
@@ -86,8 +89,6 @@ class MemDepUnit
     std::string _name;
 
   public:
-    typedef typename Impl::O3CPU O3CPU;
-
     /** Empty constructor. Must call init() prior to using in this case. */
     MemDepUnit();
 
@@ -101,7 +102,8 @@ class MemDepUnit
     std::string name() const { return _name; }
 
     /** Initializes the unit with parameters and a thread id. */
-    void init(const DerivO3CPUParams &params, ThreadID tid, O3CPU *cpu);
+    void init(const DerivO3CPUParams &params, ThreadID tid,
+            FullO3CPU<Impl> *cpu);
 
     /** Determine if we are drained. */
     bool isDrained() const;
diff --git a/src/cpu/o3/mem_dep_unit_impl.hh b/src/cpu/o3/mem_dep_unit_impl.hh
index 34bba53d95..ce1aa2639b 100644
--- a/src/cpu/o3/mem_dep_unit_impl.hh
+++ b/src/cpu/o3/mem_dep_unit_impl.hh
@@ -99,7 +99,7 @@ MemDepUnit<MemDepPred, Impl>::~MemDepUnit()
 template <class MemDepPred, class Impl>
 void
 MemDepUnit<MemDepPred, Impl>::init(
-        const DerivO3CPUParams &params, ThreadID tid, O3CPU *cpu)
+        const DerivO3CPUParams &params, ThreadID tid, FullO3CPU<Impl> *cpu)
 {
     DPRINTF(MemDepUnit, "Creating MemDepUnit %i object.\n",tid);
 
diff --git a/src/cpu/o3/probe/elastic_trace.hh b/src/cpu/o3/probe/elastic_trace.hh
index 70f2763b8e..914e4eab4e 100644
--- a/src/cpu/o3/probe/elastic_trace.hh
+++ b/src/cpu/o3/probe/elastic_trace.hh
@@ -60,6 +60,9 @@
 #include "sim/eventq.hh"
 #include "sim/probe/probe.hh"
 
+template <class Impl>
+class FullO3CPU;
+
 /**
  * The elastic trace is a type of probe listener and listens to probe points
  * in multiple stages of the O3CPU. The notify method is called on a probe
diff --git a/src/cpu/o3/rename.hh b/src/cpu/o3/rename.hh
index 2c4796a0bd..0fa8f18f20 100644
--- a/src/cpu/o3/rename.hh
+++ b/src/cpu/o3/rename.hh
@@ -74,7 +74,6 @@ class DefaultRename
 {
   public:
     // Typedefs from the Impl.
-    typedef typename Impl::O3CPU O3CPU;
     typedef typename Impl::DecodeStruct DecodeStruct;
     typedef typename Impl::RenameStruct RenameStruct;
     typedef typename Impl::TimeStruct TimeStruct;
@@ -126,7 +125,7 @@ class DefaultRename
 
   public:
     /** DefaultRename constructor. */
-    DefaultRename(O3CPU *_cpu, const DerivO3CPUParams &params);
+    DefaultRename(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     /** Returns the name of rename. */
     std::string name() const;
@@ -320,7 +319,7 @@ class DefaultRename
     std::list<RenameHistory> historyBuffer[O3MaxThreads];
 
     /** Pointer to CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Pointer to main time buffer used for backwards communication. */
     TimeBuffer<TimeStruct> *timeBuffer;
diff --git a/src/cpu/o3/rename_impl.hh b/src/cpu/o3/rename_impl.hh
index bc33c5d83d..1a5409ba68 100644
--- a/src/cpu/o3/rename_impl.hh
+++ b/src/cpu/o3/rename_impl.hh
@@ -53,7 +53,8 @@
 #include "params/DerivO3CPU.hh"
 
 template <class Impl>
-DefaultRename<Impl>::DefaultRename(O3CPU *_cpu, const DerivO3CPUParams &params)
+DefaultRename<Impl>::DefaultRename(FullO3CPU<Impl> *_cpu,
+        const DerivO3CPUParams &params)
     : cpu(_cpu),
       iewToRenameDelay(params.iewToRenameDelay),
       decodeToRenameDelay(params.decodeToRenameDelay),
@@ -865,7 +866,7 @@ DefaultRename<Impl>::updateStatus()
 
             DPRINTF(Activity, "Activating stage.\n");
 
-            cpu->activateStage(O3CPU::RenameIdx);
+            cpu->activateStage(FullO3CPU<Impl>::RenameIdx);
         }
     } else {
         // If it's not unblocking, then rename will not have any internal
@@ -874,7 +875,7 @@ DefaultRename<Impl>::updateStatus()
             _status = Inactive;
             DPRINTF(Activity, "Deactivating stage.\n");
 
-            cpu->deactivateStage(O3CPU::RenameIdx);
+            cpu->deactivateStage(FullO3CPU<Impl>::RenameIdx);
         }
     }
 }
diff --git a/src/cpu/o3/rob.hh b/src/cpu/o3/rob.hh
index 1259b53f8d..155352a17d 100644
--- a/src/cpu/o3/rob.hh
+++ b/src/cpu/o3/rob.hh
@@ -58,9 +58,6 @@ template <class Impl>
 class ROB
 {
   public:
-    //Typedefs from the Impl.
-    typedef typename Impl::O3CPU O3CPU;
-
     typedef std::pair<RegIndex, RegIndex> UnmapInfo;
     typedef typename std::list<O3DynInstPtr>::iterator InstIt;
 
@@ -84,7 +81,7 @@ class ROB
      *  @param _cpu   The cpu object pointer.
      *  @param params The cpu params including several ROB-specific parameters.
      */
-    ROB(O3CPU *_cpu, const DerivO3CPUParams &params);
+    ROB(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params);
 
     std::string name() const;
 
@@ -261,7 +258,7 @@ class ROB
     void resetState();
 
     /** Pointer to the CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     /** Active Threads in CPU */
     std::list<ThreadID> *activeThreads;
diff --git a/src/cpu/o3/rob_impl.hh b/src/cpu/o3/rob_impl.hh
index 6bdf23a790..c48484c091 100644
--- a/src/cpu/o3/rob_impl.hh
+++ b/src/cpu/o3/rob_impl.hh
@@ -51,7 +51,7 @@
 #include "params/DerivO3CPU.hh"
 
 template <class Impl>
-ROB<Impl>::ROB(O3CPU *_cpu, const DerivO3CPUParams &params)
+ROB<Impl>::ROB(FullO3CPU<Impl> *_cpu, const DerivO3CPUParams &params)
     : robPolicy(params.smtROBPolicy),
       cpu(_cpu),
       numEntries(params.numROBEntries),
diff --git a/src/cpu/o3/thread_context.hh b/src/cpu/o3/thread_context.hh
index 8dd4165bb7..5a63641aa5 100644
--- a/src/cpu/o3/thread_context.hh
+++ b/src/cpu/o3/thread_context.hh
@@ -63,10 +63,8 @@ template <class Impl>
 class O3ThreadContext : public ThreadContext
 {
   public:
-    typedef typename Impl::O3CPU O3CPU;
-
    /** Pointer to the CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
     bool
     schedule(PCEvent *e) override
diff --git a/src/cpu/o3/thread_state.hh b/src/cpu/o3/thread_state.hh
index 03091c7729..68d689ebe1 100644
--- a/src/cpu/o3/thread_state.hh
+++ b/src/cpu/o3/thread_state.hh
@@ -64,11 +64,10 @@ template <class Impl>
 struct O3ThreadState : public ThreadState
 {
     typedef ThreadContext::Status Status;
-    typedef typename Impl::O3CPU O3CPU;
 
   private:
     /** Pointer to the CPU. */
-    O3CPU *cpu;
+    FullO3CPU<Impl> *cpu;
 
   public:
     PCEventQueue pcEventQueue;
@@ -96,7 +95,7 @@ struct O3ThreadState : public ThreadState
     /** Pointer to the hardware transactional memory checkpoint. */
     std::unique_ptr<BaseHTMCheckpoint> htmCheckpoint;
 
-    O3ThreadState(O3CPU *_cpu, int _thread_num, Process *_process)
+    O3ThreadState(FullO3CPU<Impl> *_cpu, int _thread_num, Process *_process)
         : ThreadState(_cpu, _thread_num, _process), cpu(_cpu),
           comInstEventQueue("instruction-based event queue"),
           noSquashFromTC(false), trapPending(false), tc(nullptr)